Open almost any seed deck with 'AI' in the first line and the price tag looks different from the company next to it. The round is larger, the valuation sits above the benchmark, and the founder has options. UK angels keep asking the same thing about it: is this durable demand, or is it heat that unwinds the moment the music slows? In short, is AI seed valuation a bubble?
This page won't answer that with a verdict, because the honest answer is that serious people disagree and the data cuts both ways. What it will do is lay out the evidence on each side, every figure attributed to a named source, so you can read the round in front of you with a clearer head. If you want the non-AI half of the same story, the split between hot and cold sectors, that's the companion piece on the two-tier seed market.
The price is a fact. Whether it's justified is the argument, and it runs through named voices on both sides.
Why angels are asking this in 2026
Angels ask now because AI rounds visibly price and raise above the norm, and nobody is sure the demand behind them is built to last. That tension is the whole question. An AI seed company can command a larger cheque and a richer valuation than a comparable non-AI business at the same stage, and the angel writing into it has to decide whether that gap reflects real, sticky revenue or a crowd bidding the same theme.
It matters at the cheque level, not just the headline. When a theme runs hot, the discipline that normally anchors a seed price, the benchmarks, the dilution, the milestone the money is meant to buy, can quietly loosen. A round priced for a story is the classic way an angel overpays. The flip side is just as real: walk away from genuine demand because it looks frothy, and you miss the businesses that justify the price. Neither reflex is safe alone, which is why the evidence on both sides is worth reading before you lean.
The heat, with the numbers
The heat is real and measurable. UK AI deals ran about 40% larger by size in 2024, an £8.3m average against £5.7m market-wide, per the British Business Bank's Small Business Equity Tracker 2025. Capital is also concentrating hard: Carta put roughly half of US pre-seed dollars into AI by Q1 2026. More money, larger rounds, one theme.
Note what the Bank figure is and isn't. It measures round size, not a pure valuation premium, so the right phrasing is that AI rounds run about 40% larger, not that AI prices 40% higher. For the pure premium the only clean reads are vendor and global, not UK: PitchBook reported a 43.9% AI seed premium in Q4 2025 (an e-commerce-SaaS subset), and Carta put the US seed median post-money at an all-time high of $24m in the same quarter. Treat those as direction, attributed, and labelled global.
At the top of the market the concentration is starker still. KPMG's Venture Pulse for Q1 2026 recorded a record $330.9bn of global VC, driven by AI megadeals: OpenAI's $122bn raise, Anthropic's $30.6bn, xAI's $20bn. Those are growth-stage numbers, far from a seed cheque, but they set the field the whole asset class prices inside.
The bull case, named
The case for the prices being earned, not imagined, rests on revenue arriving faster than the doubters predicted. Anthropic reached roughly $30bn in annualised revenue by April 2026, most of it enterprise rather than consumer, according to company figures reported in the press. That is real money paid by businesses for a product, not a projection on a slide.
The smart-money signal points the same way. Sequoia, one of the funds closest to this trade, kept raising new vehicles rather than pulling back, which is not what a firm does when it thinks the floor is about to go. And the analyst Timothy B. Lee has argued the revenue is materialising sooner and larger than bubble-sceptics expected. That is the crux of the bull read: if the cash flows show up early, a high price can be a fair price for a fast-compounding business.
None of that proves every AI seed round is sensibly valued. It does mean the dismissal 'it's all hype, there's no revenue' no longer holds at the top of the market. The demand is partly there, named and paid for, and a fair read of the bubble question has to start from that.
The bear case, named
The caution comes from inside the industry, not just from outside critics. Sequoia's David Cahn framed a roughly $600bn-a-year gap (December 2025) between the revenue AI companies are generating and the infrastructure spend being laid down to serve them. His point isn't that the technology fails; it's that the spending has run ahead of the cash flows that are supposed to pay for it, and that gap has to close somehow.
The operators are blunt too. OpenAI's Sam Altman said in autumn 2025 that investors are 'overexcited' about AI and that 'people will overinvest and lose money', which is a striking thing to hear from the chief executive at the centre of the boom. And on whether the technology is paying its way inside companies yet, MIT's Project NANDA reported in July 2025 that 95% of enterprise generative-AI pilots showed no measurable impact on profit and loss. Adoption is one thing; a return on it is another, and the early enterprise evidence is thin.
Put those together and the bear read is not that AI does nothing. It is that price and spend may have outrun proven, durable revenue, and that some of the capital going in at today's valuations will not come back. That is a coherent position, held by people who are long the sector, which is what makes it worth taking seriously.
What this means for the round in front of you, and what it isn't
You can hold the question open and still make a decision. Nothing above tells you AI is cheap or dear, and this page deliberately won't: that would be a market-timing call, and it isn't ours to make. What the evidence does is set the field. The heat is real, the concentration is real, and serious named voices both warn about it and reassure about it.
So the work comes back to the single company on your desk. Is this round priced sanely against the benchmarks, the stage and the dilution, or is it carrying a sector premium it can't justify on its own? That sanity-check method lives in the valuation sanity-check, the UK number ranges to check it against are in the 2026 seed valuation benchmarks, and the non-AI half of the split is in the two-tier seed market. A theme being hot is information, not a verdict on the business.
To be plain: this is general information, not financial advice, and it is not a steer for or against AI as a category. Early-stage AI companies carry the same brutal base rate of failure as any other startup, and a high entry price raises the bar for a return, whatever the theme. Price each company on its own evidence, confirm any figure here at the named source on the day you read it, and take FCA-regulated advice before you commit capital.
Frequently asked questions
Is AI in a bubble in 2026?
There is a real case on both sides, and The Carry doesn't make the call. The data shows clear heat and concentration: the British Business Bank found UK AI deals about 40% larger by size in 2024, and Carta put roughly half of US pre-seed dollars into AI by Q1 2026. Named voices warn (Sequoia's David Cahn on a roughly $600bn revenue gap, Sam Altman calling investors overexcited, MIT's NANDA finding 95% of GenAI pilots showed no profit impact) and reassure (Anthropic at about $30bn annualised revenue, Sequoia still raising). This is general information, not advice.
Why are AI valuations so high?
Because AI rounds are larger and capital is heavily concentrated in the theme. The British Business Bank found UK AI deals about 40% larger by deal size in 2024 (£8.3m average against £5.7m market-wide), and Carta reported roughly half of US pre-seed dollars going to AI by Q1 2026. Whether that pricing is justified by durable demand or inflated by a crowded trade is exactly the debate, and it isn't settled.
Are non-AI startups cheaper?
The seed market has split, with AI-flavoured rounds pricing and raising above the rest. That divergence is the subject of the companion piece on the two-tier seed market, which covers the non-AI side. For the UK number ranges a non-AI round is checked against, see the 2026 seed valuation benchmarks.
Should I avoid AI startups?
That is an investment decision, and this page won't make it for you. Whether AI is overpriced, fairly priced or cheap is a market-timing call The Carry doesn't make. Price each company on its own merits against the benchmarks and the stage, rather than backing or avoiding a whole theme. The data and the named arguments above are there to inform that judgement, not to replace it.
Where can I check these figures and get proper advice?
Every figure here is attributed to a named source you can confirm directly: the British Business Bank's Equity Tracker, Carta and PitchBook for the vendor reads, KPMG's Venture Pulse, and the public statements from Anthropic, OpenAI, Sequoia and MIT's Project NANDA. Market data moves, so re-check the numbers you lean on at the source on the day you read this. For your own situation, take advice from an FCA-regulated adviser; this article is general information only.